Linearization means converting RAW sensor data into values that are proportional to the actual light captured.

In an ideal linear RAW file, doubling the light would double the recorded value. In practice, that may not be true straight out of the camera because of sensor response quirks, RAW compression/tone encoding, highlight handling, or shadow/flare effects. So a RAW converter may first apply a mapping or lookup table to restore a linear relationship.

This matters because later RAW-processing steps usually assume linear data. Operations such as demosaicing and white-balance scaling work correctly only if the input values represent light intensity in a linear way; otherwise they can introduce errors.

From the community answers, “linearization” can include:

• undoing tone-compressed RAW encoding using metadata tables,
• correcting nonlinear behavior near clipped highlights,
• compensating for shadow nonlinearity caused by flare or lens-dependent effects.

So, in short: linearization is an early RAW-processing step that makes the recorded sensor values behave like true scene-light measurements before further processing.